Live Updates > Live Update Details
2025-09-19 09:17:18
[Xiaomi open sources its first native end-to-end speech model] On September 19, Xiaomi officially open-sourced its first native end-to-end speech model - Xiaomi-MiMo-Audio. Based on an innovative pre-training architecture and hundreds of millions of hours of training data, it achieved ICL-based few-sample generalization in the speech field for the first time, and observed obvious "emergence" behavior in pre-training. It is reported that in a number of standard evaluation benchmarks such as general speech understanding and dialogue, MiMo-Audio significantly surpassed the open source model with the same number of parameters, achieving the best performance of 7B; on the standard test set of the audio understanding benchmark MMAU, MiMo-Audio surpassed Google's closed-source speech model Gemini-2.5-Flash; in the Big Bench Audio S2T task, a benchmark for complex audio reasoning, MiMo-Audio also surpassed OpenAI's closed-source speech model GPT-4o-Audio-Preview
